Recognition-by-components theory
Recognition-by-components (RBC) theory, proposed by Irving Biederman in 1987, explains how we recognize objects. The idea is that we break objects down into a small set of basic 3D shapes called geons (like cylinders, cones, boxes, and other simple forms). There are fewer than 36 geons, and objects are built by combining a few of them. For example, a mug can be recognized as a cylinder for the body plus a handle.
We identify objects by comparing the arrangement of geons we see with what we have stored in memory. Biederman compares this to language: English uses about 44 phonemes, and all languages use about 55; similarly, a finite set of geons can compose many different objects.
A key idea is viewpoint invariance: we can recognize objects from many different angles because geons have edge properties that stay recognizable as we look from different viewpoints. This gives two main advantages. First, a single geon description can represent an object from all angles. Second, a small set of geons forms an “alphabet” that can combine to create many objects. For example, with 24 geons, there could be billions of possible 3-geon combinations.
Some evidence even suggests that infants as young as four months may start to recognize geons and their combinations, hinting that this is a fundamental perceptual skill.
RBC has limits. It doesn’t explain how to turn a real photograph or scene into a geon-based description, and geons plus their relationships can fail to distinguish some objects (like a pear vs. an apple) that look similar. Biederman argues RBC is the primary way we recognize most objects, with a secondary process handling cases where geons aren’t enough.
This page was last edited on 3 February 2026, at 05:18 (CET).